Definition : If a n by n matrix $A$ is orthogonally congruent to another matrix $B$, then there exist an orthogonal matrix $C$ such that $$A = C^{-1}BC$$

Theorem: If $A$ is symmetric, then $A$ is orthogonally congruent to a diagonal matrix, $B = diag (\lambda_1 , .....,\lambda_n)$ where $\lambda_1,....\lambda_n$ are the (necessarily real) eigenvalues of $A$ (with multiple eigenvalues repeated in accordance with their multiplicities)

Why is this true?? If $A$ has n distinct eigenvalues its obvious that this is true, for we can have, $$C = col (\bf{v_1},v_2,....v_n)$$ where $\bf{v_1,v_2...,v_n}$ are all the eigenvectors associated with the eigenvalues, and $$ AC = CB \\ B = diag(\lambda_1,....\lambda_n) \\ A = C^{-1}BC$$ But if we do the same thing for the case where the eigenvalues are repeated, then wouldnt $C$ have multiple columns of the linearly dependent vectors? (If we use the repeated eigenvalues to solve for the eigenvectors wouldnt they be the same?) If so, $det(C) = 0$ then how isit still possible that $$A = C^{-1}BC$$ when $C^{-1}$ doesnt exist??

  • $\begingroup$ Basically, If $A$ is symmetric and $v$ is an eigenvector of $A$, then we have that the orthogonal complement to $\left<v\right>$ is invariant under multiplication by $A$. Check out page 17 here: maths.ox.ac.uk/system/files/coursematerial/2013/2632/18/… $\endgroup$
    – JC574
    Jul 23 '14 at 15:09
  • 1
    $\begingroup$ Dear CodeBreaker: I retitled the question so that people would not automatically ignore it, and so that there would be a remote chance someone with a similar question would be able to find it. Please consider taking the same things into consideration when titling your next question! Regards $\endgroup$
    – rschwieb
    Jul 23 '14 at 15:17

There's a few points you're missing here:

  • Just because we have $n$ linearly independent vectors, doesn't mean that they form an orthonormal basis. If $A$ were any matrix with $n$ distinct eigenvalues, then $C = col(v_1, \dots, v_n)$ would have to be invertible, but is not necessarily orthogonal.
  • Some matrices have eigenspaces that are more than $1$-dimensional. As an example, $\pmatrix{1&0&0\\0&1&0\\0&0&1}$ has only one eigenvalue, but certainly has a basis of eigenvectors.

Hopefully that clears some things up. This, of course, is not a proof, but presumably this is proved in your textbook.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.